ganttpro-mcp

A local Model Context Protocol (MCP) server that exposes the full app.ganttpro.com REST API (https://api.ganttpro.com/v1.0) as MCP tools. Runs locally over stdio — your API key and data never pass through a third party.

What it covers

One MCP tool per documented GanttPRO endpoint (43 tools) across every resource group:

| Group | Tools | |-------|-------| | Projects | list_projects, create_project, get_project, update_project, get_task_fields, get_calendars, create_custom_day, update_custom_day, delete_custom_day | | Tasks | list_tasks, create_task, get_task, update_task, delete_task, assign_resource, update_assigned_resource, unassign_resource | | Comments | list_comments, get_comments_by_project, create_comment, update_comment, delete_comment | | Links | create_link, get_link, update_link, delete_link | | Time logs | list_time_logs, get_time_logs_by_project, create_time_log, update_time_log, delete_time_log | | Resources | list_resources | | Attachments | list_attachments, create_attachment, delete_attachment | | Team / Users | get_team, list_users, get_user, update_user | | Roles / Reference | get_account_roles, get_project_roles, get_languages, get_colors |

All tools are prefixed ganttpro_ (e.g. ganttpro_create_task).

Write endpoints (create_*/update_*) accept any extra body field beyond the documented ones (additionalProperties: true), so the server keeps working if GanttPRO adds new fields.

Quick start (no clone, no install)

The only prerequisite is Node.js (LTS). The server runs straight from npm via npx — no clone, no npm install, no paths:

# verify your API key works
GANTTPRO_API_KEY=xxxx npx -y @redcollar/ganttpro-mcp check

# print a ready-to-paste config for your client
npx -y @redcollar/ganttpro-mcp config claude     # or: cursor | cherry | librechat | (none = all)

Get your API key from app.ganttpro.com → Account settings → API. It is sent on every request as the X-API-KEY header.

CLI commands

| Command | Purpose | |---------|---------| | npx -y @redcollar/ganttpro-mcp | Run the MCP server on stdio (how clients launch it) | | npx -y @redcollar/ganttpro-mcp check | Verify the API key (lists projects, prints ✅/❌) | | npx -y @redcollar/ganttpro-mcp config [client] | Print a ready-to-paste config (claude, cursor, cherry, librechat) | | npx -y @redcollar/ganttpro-mcp help | Usage |

Environment variables

| Var | Required | Default | Purpose | |-----|----------|---------|---------| | GANTTPRO_API_KEY | ✅ | — | Your GanttPRO API key | | GANTTPRO_BASE_URL | — | https://api.ganttpro.com/v1.0 | API base URL override | | GANTTPRO_ARRAY_FORMAT | — | brackets | Array/object query serialization: brackets (id[]=1), repeat (id=1&id=2), or comma (id=1,2) |

Add to a client

Paste this into your client config (Claude Desktop claude_desktop_config.json, Cursor ~/.cursor/mcp.json, or .mcp.json):

{
  "mcpServers": {
    "ganttpro": {
      "command": "npx",
      "args": ["-y", "@redcollar/ganttpro-mcp"],
      "env": { "GANTTPRO_API_KEY": "your-api-key-here" }
    }
  }
}

Or, in Claude Code:

claude mcp add ganttpro --env GANTTPRO_API_KEY=your-api-key-here -- npx -y @redcollar/ganttpro-mcp

Then restart the client and check it connects.

Which clients work

Local stdio servers run in desktop apps (Claude Desktop, Cursor, Cherry Studio) and self-hosted LibreChat (admin-configured via librechat.yaml). The web ChatGPT supports MCP only over remote HTTP, and the web DeepSeek has no MCP client — for those, use a desktop app (with ChatGPT/DeepSeek as the model) and connect GanttPRO there.

Local development

cd ganttpro-mcp
npm install
npm test                              # offline unit + handshake tests (27)
GANTTPRO_API_KEY=xxxx node index.js   # run the server directly from source

Usage notes

• Rate limit: GanttPRO allows ~5 write requests/second (insert/update/delete). The server forwards calls 1:1 — batch large operations accordingly.
• Dates: pass dates as strings in the format your GanttPRO account expects (e.g. "2026-06-04").
• Filtering tasks: range filters (progress, duration, startDate, …) take objects like { "from": 0, "to": 0.5 }; list filters (status, priority, …) take arrays.
• Each tool result is HTTP <status> OK/ERROR followed by the JSON response body. Non-2xx responses come back with isError: true.

Layout

ganttpro-mcp/
├── index.js                # CLI dispatcher: server (default) / check / config / help
├── src/
│   ├── endpoints.js        # data-only catalog of every API endpoint
│   ├── client.js           # request building + fetch (unit-testable)
│   ├── tools.js            # endpoint specs -> MCP tools + dispatch
│   ├── server.js           # MCP Server wiring (ListTools / CallTool)
│   ├── cli.js              # check / config / help logic
│   └── config-snippets.js  # per-client config templates (npx-based)
└── test/
    ├── smoke.test.js       # client/tools unit tests (mocked fetch)
    ├── server.test.js      # real Client<->Server handshake
    └── cli.test.js         # check / config / help tests

License

MIT